4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/string.h>
9 #include <linux/file.h>
10 #include <linux/fdtable.h>
11 #include <linux/fsnotify.h>
12 #include <linux/module.h>
13 #include <linux/tty.h>
14 #include <linux/namei.h>
15 #include <linux/backing-dev.h>
16 #include <linux/capability.h>
17 #include <linux/securebits.h>
18 #include <linux/security.h>
19 #include <linux/mount.h>
20 #include <linux/fcntl.h>
21 #include <linux/slab.h>
22 #include <asm/uaccess.h>
24 #include <linux/personality.h>
25 #include <linux/pagemap.h>
26 #include <linux/syscalls.h>
27 #include <linux/rcupdate.h>
28 #include <linux/audit.h>
29 #include <linux/falloc.h>
30 #include <linux/fs_struct.h>
31 #include <linux/ima.h>
32 #include <linux/dnotify.h>
36 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
40 struct iattr newattrs;
42 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 newattrs.ia_size = length;
47 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 newattrs.ia_file = filp;
50 newattrs.ia_valid |= ATTR_FILE;
53 /* Remove suid/sgid on truncate too */
54 ret = should_remove_suid(dentry);
56 newattrs.ia_valid |= ret | ATTR_FORCE;
58 mutex_lock(&dentry->d_inode->i_mutex);
59 ret = notify_change(dentry, &newattrs);
60 mutex_unlock(&dentry->d_inode->i_mutex);
64 static long do_sys_truncate(const char __user *pathname, loff_t length)
71 if (length < 0) /* sorry, but loff_t says... */
74 error = user_path(pathname, &path);
77 inode = path.dentry->d_inode;
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
81 if (S_ISDIR(inode->i_mode))
85 if (!S_ISREG(inode->i_mode))
88 error = mnt_want_write(path.mnt);
92 error = inode_permission(inode, MAY_WRITE);
94 goto mnt_drop_write_and_out;
98 goto mnt_drop_write_and_out;
100 error = get_write_access(inode);
102 goto mnt_drop_write_and_out;
105 * Make sure that there are no leases. get_write_access() protects
106 * against the truncate racing with a lease-granting setlease().
108 error = break_lease(inode, O_WRONLY);
110 goto put_write_and_out;
112 error = locks_verify_truncate(inode, NULL, length);
114 error = security_path_truncate(&path, length, 0);
116 error = do_truncate(path.dentry, length, 0, NULL);
119 put_write_access(inode);
120 mnt_drop_write_and_out:
121 mnt_drop_write(path.mnt);
128 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
130 return do_sys_truncate(path, length);
133 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
135 struct inode * inode;
136 struct dentry *dentry;
148 /* explicitly opened as large or we are on 64-bit box */
149 if (file->f_flags & O_LARGEFILE)
152 dentry = file->f_path.dentry;
153 inode = dentry->d_inode;
155 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
159 /* Cannot ftruncate over 2^31 bytes without large file support */
160 if (small && length > MAX_NON_LFS)
164 if (IS_APPEND(inode))
167 error = locks_verify_truncate(inode, file, length);
169 error = security_path_truncate(&file->f_path, length,
170 ATTR_MTIME|ATTR_CTIME);
172 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
179 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
181 long ret = do_sys_ftruncate(fd, length, 1);
182 /* avoid REGPARM breakage on x86: */
183 asmlinkage_protect(2, ret, fd, length);
187 /* LFS versions of truncate are only needed on 32 bit machines */
188 #if BITS_PER_LONG == 32
189 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
191 return do_sys_truncate(path, length);
193 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
194 asmlinkage long SyS_truncate64(long path, loff_t length)
196 return SYSC_truncate64((const char __user *) path, length);
198 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
201 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
203 long ret = do_sys_ftruncate(fd, length, 0);
204 /* avoid REGPARM breakage on x86: */
205 asmlinkage_protect(2, ret, fd, length);
208 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
209 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
211 return SYSC_ftruncate64((unsigned int) fd, length);
213 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
215 #endif /* BITS_PER_LONG == 32 */
218 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
220 struct inode *inode = file->f_path.dentry->d_inode;
223 if (offset < 0 || len <= 0)
226 /* Return error if mode is not supported */
227 if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
230 if (!(file->f_mode & FMODE_WRITE))
233 * Revalidate the write permissions, in case security policy has
234 * changed since the files were opened.
236 ret = security_file_permission(file, MAY_WRITE);
240 if (S_ISFIFO(inode->i_mode))
244 * Let individual file system decide if it supports preallocation
245 * for directories or not.
247 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
250 /* Check for wrap through zero too */
251 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
254 if (!inode->i_op->fallocate)
257 return inode->i_op->fallocate(inode, mode, offset, len);
260 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
267 error = do_fallocate(file, mode, offset, len);
274 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
275 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
277 return SYSC_fallocate((int)fd, (int)mode, offset, len);
279 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
283 * access() needs to use the real uid/gid, not the effective uid/gid.
284 * We do this by temporarily clearing all FS-related capabilities and
285 * switching the fsuid/fsgid around to the real ones.
287 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
289 const struct cred *old_cred;
290 struct cred *override_cred;
295 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
298 override_cred = prepare_creds();
302 override_cred->fsuid = override_cred->uid;
303 override_cred->fsgid = override_cred->gid;
305 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
306 /* Clear the capabilities if we switch to a non-root user */
307 if (override_cred->uid)
308 cap_clear(override_cred->cap_effective);
310 override_cred->cap_effective =
311 override_cred->cap_permitted;
314 old_cred = override_creds(override_cred);
316 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
320 inode = path.dentry->d_inode;
322 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
324 * MAY_EXEC on regular files is denied if the fs is mounted
325 * with the "noexec" flag.
328 if (path.mnt->mnt_flags & MNT_NOEXEC)
329 goto out_path_release;
332 res = inode_permission(inode, mode | MAY_ACCESS);
333 /* SuS v2 requires we report a read only fs too */
334 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
335 goto out_path_release;
337 * This is a rare case where using __mnt_is_readonly()
338 * is OK without a mnt_want/drop_write() pair. Since
339 * no actual write to the fs is performed here, we do
340 * not need to telegraph to that to anyone.
342 * By doing this, we accept that this access is
343 * inherently racy and know that the fs may change
344 * state before we even see this result.
346 if (__mnt_is_readonly(path.mnt))
352 revert_creds(old_cred);
353 put_cred(override_cred);
357 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
359 return sys_faccessat(AT_FDCWD, filename, mode);
362 SYSCALL_DEFINE1(chdir, const char __user *, filename)
367 error = user_path_dir(filename, &path);
371 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
375 set_fs_pwd(current->fs, &path);
383 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
394 inode = file->f_path.dentry->d_inode;
397 if (!S_ISDIR(inode->i_mode))
400 error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
402 set_fs_pwd(current->fs, &file->f_path);
409 SYSCALL_DEFINE1(chroot, const char __user *, filename)
414 error = user_path_dir(filename, &path);
418 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
423 if (!capable(CAP_SYS_CHROOT))
425 error = security_path_chroot(&path);
429 set_fs_root(current->fs, &path);
437 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
439 struct inode * inode;
440 struct dentry * dentry;
443 struct iattr newattrs;
449 dentry = file->f_path.dentry;
450 inode = dentry->d_inode;
452 audit_inode(NULL, dentry);
454 err = mnt_want_write_file(file);
457 mutex_lock(&inode->i_mutex);
458 err = security_path_chmod(dentry, file->f_vfsmnt, mode);
461 if (mode == (mode_t) -1)
462 mode = inode->i_mode;
463 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
464 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
465 err = notify_change(dentry, &newattrs);
467 mutex_unlock(&inode->i_mutex);
468 mnt_drop_write(file->f_path.mnt);
475 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
480 struct iattr newattrs;
482 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
485 inode = path.dentry->d_inode;
487 error = mnt_want_write(path.mnt);
490 mutex_lock(&inode->i_mutex);
491 error = security_path_chmod(path.dentry, path.mnt, mode);
494 if (mode == (mode_t) -1)
495 mode = inode->i_mode;
496 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
497 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
498 error = notify_change(path.dentry, &newattrs);
500 mutex_unlock(&inode->i_mutex);
501 mnt_drop_write(path.mnt);
508 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
510 return sys_fchmodat(AT_FDCWD, filename, mode);
513 static int chown_common(struct path *path, uid_t user, gid_t group)
515 struct inode *inode = path->dentry->d_inode;
517 struct iattr newattrs;
519 newattrs.ia_valid = ATTR_CTIME;
520 if (user != (uid_t) -1) {
521 newattrs.ia_valid |= ATTR_UID;
522 newattrs.ia_uid = user;
524 if (group != (gid_t) -1) {
525 newattrs.ia_valid |= ATTR_GID;
526 newattrs.ia_gid = group;
528 if (!S_ISDIR(inode->i_mode))
530 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
531 mutex_lock(&inode->i_mutex);
532 error = security_path_chown(path, user, group);
534 error = notify_change(path->dentry, &newattrs);
535 mutex_unlock(&inode->i_mutex);
540 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
545 error = user_path(filename, &path);
548 error = mnt_want_write(path.mnt);
551 error = chown_common(&path, user, group);
552 mnt_drop_write(path.mnt);
559 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
560 gid_t, group, int, flag)
566 if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
569 follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
570 error = user_path_at(dfd, filename, follow, &path);
573 error = mnt_want_write(path.mnt);
576 error = chown_common(&path, user, group);
577 mnt_drop_write(path.mnt);
584 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
589 error = user_lpath(filename, &path);
592 error = mnt_want_write(path.mnt);
595 error = chown_common(&path, user, group);
596 mnt_drop_write(path.mnt);
603 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
607 struct dentry * dentry;
613 error = mnt_want_write_file(file);
616 dentry = file->f_path.dentry;
617 audit_inode(NULL, dentry);
618 error = chown_common(&file->f_path, user, group);
619 mnt_drop_write(file->f_path.mnt);
627 * You have to be very careful that these write
628 * counts get cleaned up in error cases and
629 * upon __fput(). This should probably never
630 * be called outside of __dentry_open().
632 static inline int __get_file_write_access(struct inode *inode,
633 struct vfsmount *mnt)
636 error = get_write_access(inode);
640 * Do not take mount writer counts on
641 * special files since no writes to
642 * the mount itself will occur.
644 if (!special_file(inode->i_mode)) {
646 * Balanced in __fput()
648 error = mnt_want_write(mnt);
650 put_write_access(inode);
655 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
657 int (*open)(struct inode *, struct file *),
658 const struct cred *cred)
663 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
664 FMODE_PREAD | FMODE_PWRITE;
665 inode = dentry->d_inode;
666 if (f->f_mode & FMODE_WRITE) {
667 error = __get_file_write_access(inode, mnt);
670 if (!special_file(inode->i_mode))
674 f->f_mapping = inode->i_mapping;
675 f->f_path.dentry = dentry;
678 f->f_op = fops_get(inode->i_fop);
679 file_move(f, &inode->i_sb->s_files);
681 error = security_dentry_open(f, cred);
685 if (!open && f->f_op)
686 open = f->f_op->open;
688 error = open(inode, f);
694 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
696 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
698 /* NB: we're sure to have correct a_ops only after f_op->open */
699 if (f->f_flags & O_DIRECT) {
700 if (!f->f_mapping->a_ops ||
701 ((!f->f_mapping->a_ops->direct_IO) &&
702 (!f->f_mapping->a_ops->get_xip_mem))) {
704 f = ERR_PTR(-EINVAL);
712 if (f->f_mode & FMODE_WRITE) {
713 put_write_access(inode);
714 if (!special_file(inode->i_mode)) {
716 * We don't consider this a real
717 * mnt_want/drop_write() pair
718 * because it all happenend right
719 * here, so just reset the state.
726 f->f_path.dentry = NULL;
727 f->f_path.mnt = NULL;
732 return ERR_PTR(error);
736 * lookup_instantiate_filp - instantiates the open intent filp
737 * @nd: pointer to nameidata
738 * @dentry: pointer to dentry
739 * @open: open callback
741 * Helper for filesystems that want to use lookup open intents and pass back
742 * a fully instantiated struct file to the caller.
743 * This function is meant to be called from within a filesystem's
745 * Beware of calling it for non-regular files! Those ->open methods might block
746 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
747 * leading to a deadlock, as nobody can open that fifo anymore, because
748 * another process to open fifo will block on locked parent when doing lookup).
749 * Note that in case of error, nd->intent.open.file is destroyed, but the
750 * path information remains valid.
751 * If the open callback is set to NULL, then the standard f_op->open()
752 * filesystem callback is substituted.
754 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
755 int (*open)(struct inode *, struct file *))
757 const struct cred *cred = current_cred();
759 if (IS_ERR(nd->intent.open.file))
763 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
764 nd->intent.open.file,
767 return nd->intent.open.file;
769 release_open_intent(nd);
770 nd->intent.open.file = (struct file *)dentry;
773 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
776 * nameidata_to_filp - convert a nameidata to an open filp.
777 * @nd: pointer to nameidata
780 * Note that this function destroys the original nameidata
782 struct file *nameidata_to_filp(struct nameidata *nd)
784 const struct cred *cred = current_cred();
787 /* Pick up the filp from the open intent */
788 filp = nd->intent.open.file;
789 /* Has the filesystem initialised the file for us? */
790 if (filp->f_path.dentry == NULL)
791 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
799 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
802 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
803 const struct cred *cred)
808 validate_creds(cred);
811 * We must always pass in a valid mount pointer. Historically
812 * callers got away with not passing it, but we must enforce this at
813 * the earliest possible point now to avoid strange problems deep in the
817 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
819 return ERR_PTR(-EINVAL);
823 f = get_empty_filp();
827 return ERR_PTR(error);
831 return __dentry_open(dentry, mnt, f, NULL, cred);
833 EXPORT_SYMBOL(dentry_open);
835 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
837 struct fdtable *fdt = files_fdtable(files);
838 __FD_CLR(fd, fdt->open_fds);
839 if (fd < files->next_fd)
843 void put_unused_fd(unsigned int fd)
845 struct files_struct *files = current->files;
846 spin_lock(&files->file_lock);
847 __put_unused_fd(files, fd);
848 spin_unlock(&files->file_lock);
851 EXPORT_SYMBOL(put_unused_fd);
854 * Install a file pointer in the fd array.
856 * The VFS is full of places where we drop the files lock between
857 * setting the open_fds bitmap and installing the file in the file
858 * array. At any such point, we are vulnerable to a dup2() race
859 * installing a file in the array before us. We need to detect this and
860 * fput() the struct file we are about to overwrite in this case.
862 * It should never happen - if we allow dup2() do it, _really_ bad things
866 void fd_install(unsigned int fd, struct file *file)
868 struct files_struct *files = current->files;
870 spin_lock(&files->file_lock);
871 fdt = files_fdtable(files);
872 BUG_ON(fdt->fd[fd] != NULL);
873 rcu_assign_pointer(fdt->fd[fd], file);
874 spin_unlock(&files->file_lock);
877 EXPORT_SYMBOL(fd_install);
879 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
881 char *tmp = getname(filename);
882 int fd = PTR_ERR(tmp);
885 fd = get_unused_fd_flags(flags);
887 struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
901 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
905 if (force_o_largefile())
906 flags |= O_LARGEFILE;
908 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
909 /* avoid REGPARM breakage on x86: */
910 asmlinkage_protect(3, ret, filename, flags, mode);
914 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
919 if (force_o_largefile())
920 flags |= O_LARGEFILE;
922 ret = do_sys_open(dfd, filename, flags, mode);
923 /* avoid REGPARM breakage on x86: */
924 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
931 * For backward compatibility? Maybe this should be moved
932 * into arch/i386 instead?
934 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
936 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
942 * "id" is the POSIX thread ID. We use the
943 * files pointer for this..
945 int filp_close(struct file *filp, fl_owner_t id)
949 if (!file_count(filp)) {
950 printk(KERN_ERR "VFS: Close: file count is 0\n");
954 if (filp->f_op && filp->f_op->flush)
955 retval = filp->f_op->flush(filp, id);
957 dnotify_flush(filp, id);
958 locks_remove_posix(filp, id);
963 EXPORT_SYMBOL(filp_close);
966 * Careful here! We test whether the file pointer is NULL before
967 * releasing the fd. This ensures that one clone task can't release
968 * an fd while another clone is opening it.
970 SYSCALL_DEFINE1(close, unsigned int, fd)
973 struct files_struct *files = current->files;
977 spin_lock(&files->file_lock);
978 fdt = files_fdtable(files);
979 if (fd >= fdt->max_fds)
984 rcu_assign_pointer(fdt->fd[fd], NULL);
985 FD_CLR(fd, fdt->close_on_exec);
986 __put_unused_fd(files, fd);
987 spin_unlock(&files->file_lock);
988 retval = filp_close(filp, files);
990 /* can't restart close syscall because file table entry was cleared */
991 if (unlikely(retval == -ERESTARTSYS ||
992 retval == -ERESTARTNOINTR ||
993 retval == -ERESTARTNOHAND ||
994 retval == -ERESTART_RESTARTBLOCK))
1000 spin_unlock(&files->file_lock);
1003 EXPORT_SYMBOL(sys_close);
1006 * This routine simulates a hangup on the tty, to arrange that users
1007 * are given clean terminals at login time.
1009 SYSCALL_DEFINE0(vhangup)
1011 if (capable(CAP_SYS_TTY_CONFIG)) {
1019 * Called when an inode is about to be open.
1020 * We use this to disallow opening large files on 32bit systems if
1021 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1022 * on this flag in sys_open.
1024 int generic_file_open(struct inode * inode, struct file * filp)
1026 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1031 EXPORT_SYMBOL(generic_file_open);
1034 * This is used by subsystems that don't want seekable
1037 int nonseekable_open(struct inode *inode, struct file *filp)
1039 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1043 EXPORT_SYMBOL(nonseekable_open);